Application of Weighted Fuzzy Clustering Method to Supplier Selection under E-business Yan Yang1,a, Wei-ping Yang2,b, Yao Liang3,c 1 2 Department of Industrial Engineering, Kunming University of Science and Technology, Kunming, China Department of Industrial Engineering, Kunming University of Science and Technology, Kunming, China 3 Adult Education College, Kunming University of Science and Technology, Kunming, China (axyxy_1340@163.com, bywp58@yahoo.com.cn) Abstract: In order to avoid the imbalance among evaluation index in the general fuzzy clustering, using the weighted fuzzy clustering algorithm to choose the best provider under E-business. In this paper, first,the weighted of each index is curtained by AHP, and then the weight is added to the fuzzy clustering algorithm. The F-distribution of probability statistics is used to determine the best classification number, which can help choose the best classification. Finally, comprehensive value is computer by fuzzy synthetically evaluation, through which the best supplier can be selected. Keywords: Analytic hierarchy process; Fuzzy synthetically evaluation; Weight fuzzy clustering; Provider selection I Introduction The twenty-first Century is an age of information explosion, information technology and network economy is its enormous power to promote the social economic form the profound transformation, the electronic commerce has become the core of information world and the driving force for the development of network economy. E-business and the rapid development of information technology shortens the distance between enterprises and suppliers, to promote enterprise and suppliers of information integration and sharing, changing the enterprise find supplier information and supplier transaction way, affect the transaction cost, safety and reliability. Therefore, establishing the new evaluation standard is to guarantee the successful cooperation, stability of the important premise. In the traditional research of supplier selection, typical fuzzy clustering algorithm are the method based on similarity relation and fuzzy relation, the maximum tree method based on fuzzy cam and dynamic programming etc[1-2]. But the traditional algorithm is not considered in the evaluation of various factors of the difference between the, considered in the clustering process of each factor are equivalent [3]. At the same time, because of the number of suppliers which involved in electronic commerce environment is increasing, through the algorithm step select the best suppliers, will lead to excessive amount of calculation. Therefore, on the basis of considering the above two problems, in this paper, first,the weighted of each index is curtained by AHP, and then the weight is added to the fuzzy clustering algorithm. The F-distribution of probability statistics is used to determine the best classification number, which can help choose the best classification. Finally, comprehensive value is computer by fuzzy synthetically evaluation, through which the best supplier can be selected. II Establishment of evaluation index system In the era of network economy and the rapid development of e-business environment, the evaluation of supplier information ability should be given enough attention [4-6]. this paper is based on referencing the representative index of traditional supplier selection, specifically from the bright time character, scientific and practical, flexible operation, expansibility, comprehensive system five aspects to carry on comprehensive consideration, try to establish a platform for electronic commerce supplier evaluation and selection index system as shown in table I[7]: TABLE I Index system Target layer Rule layer Informationiz ation B1 Service level B2 Supplier A Index layer Information construction investment ratio C1 Computer professionals proportion C2 Information sharing integrated ability C3 Information security C4 Industry experience C5 After-sales service satisfaction C6 Historical transaction records C7 Brand reputation C8 Recycling center processing speed C9 Rapid response ability C10 Business ability B3 Technology level B4 Enterprise development prospect B5 Cost control C11 Financial status C12 Supply capacity C13 Technology innovation ability C14 Production equipment safe operation rate C15 R&D Investment ratio C16 Equipment leading level C17 Market influence C18 New product development rate C19 Training expenditure per capita C20 Economic and technological environment C21 III Identify weight by AHP This paper uses the geometric average method to solve the largest eigenvalue λ and the corresponding characteristic vector W of comparison matrix. WA=[a1 a2 a3 a4 a5] WBi=[bi1 bi2 … bin] IV Establishment of weighted fuzzy clustering model A Data weighted standardization Set domain U=(x1,x2,…,xn ) to be classified n suppliers, each object has a 5level of evaluation index, according to the problem of the original data matrix: x11 x12 x1m x x x2 m m=5 D 21 22 xn1 xn 2 xnm In practical problems, different data generally have different dimensions. In order to make a different amount of data can be compared, usually need to make appropriate transform data. Therefore, according to the fuzzy matrix requirements for data standardization data compression to the interval [0, 1], the process requires the following transformation [8]: 1)Translation · standard deviation changes: x xk xik ik sk i =1, 2 ,3,…,n; k=1,2,3,4,5 Among them, x k 1 m 1 n xik , sk ( xik x k )2 n i 1 n i 1 After transformation, each variable of the mean value is 0, the standard deviation is 1, and eliminate the influence of dimensional, but the X is not necessarily in the interval [0,1]. 2)Translation · differential changes: xik min xik xik 1i n max xik min xik 1i n 1i n 0 xik 1 k=1,2,3,4,5 is clearly, and also eliminate the j=1,2,3,…,n; And get a fuzzy relations similar matrix R rij (1) nn C Dynamic clustering process [10] Fuzzy clustering analysis requires the establishment of fuzzy matrix is reflexive, symmetric and transitive, but according to (1) type of fuzzy matrix, is a fuzzy similarity matrix R, not necessarily is transitive, that R is not necessarily a fuzzy equivalence matrix. In order to classification, R also needs to be transformed into fuzzy equivalent matrix R*. Therefore, need to use two leveling method ( such as type (2) ) on the fuzzy similar matrix R transformation, and the transitive closure of the t R (R), t (R) that is seeking a fuzzy equivalence matrix R*,that is t(R)=R*. (2) R R 2 R 4 R 2( k 1) R 2k t R rij mm By (2) type get the fuzzy equivalent matrix t(R), t(R) in numerical arranged from big to small, λ is valued to the arrangement sequence, can form the dynamic clustering figure. D Determine the optimal class number In determining the classification number, often in dynamic clustering view, adjust the λ value in order to get the proper classification, without prior to accurately estimate the good samples should be divided into several categories. This method is often in the subjective desire to classification, and then to make λ, which leads to different people to the classification, will have the different results. Thus the proposed F-distribution to determine the optimal threshold value of λ, then according to the λ value in the view of dynamic cluster classification, finally get the optimal class number. The algorithm is as follows [11-13]: Calculation of the original data matrix to overall sample center vector, in which: influence of the dimension. Then, on the transform results 1 n (k=1,2,3,4,5) (3) x xik were weighted arithmetic, get: n i 1 x11 x12 x1m a1 0 0 x11 x12 x1m Corresponding to the classification of λ set for r, x x x 0 a 0 x x x sample number of class j is nj, sample set of class j 21 22 2 m 2 21 22 2 m Y is x1 j , x2 j , xn j ,cluster center vector of class j is j j j j 0 0 0 a m xn1 xn2 xnm xn1 xn2 xnm x x1 , x 2 ,, x n B Establishing fuzzy relationship matrix According to the traditional clustering method to determine the similarity coefficient, establish fuzzy similar matrix. To determine the similarity rij R( xi , x j ) methods mainly have the traditional cluster analysis of the similarity coefficient method, distance and angle cosine method. In this paper, using the included angle cosine method, its algorithm such as type[9]: m rij x x ik k 1 m x k 1 ik 2 jk i=1,2,3,…,n; m x k 1 2 jk j 1 nj xk nj x j ik (k=1,2,3,4,5) (4) 1 According to (3) and (4) the type of F-distribution, get (5) r F n r j 1 nj j x j x x j j 1 i 1 i 2 x n r j 2 (5) r 1 Its molecular represents the distance between two classes, denominator represents the distance between the sample within-class. Therefore, the greater the F value that the distance between the classes and class is larger, classification is better. If F F r 1, n r , 0.05 ,according to the statistical analysis of variance theory that the difference between the classes is remarkable, illustrating the classification more reasonable,If the value which satisfies the inequality F F r 1, n r is more than one, can further study the size of ( F F ) , find a satisfactory F value from larger. V Multistage fuzzy comprehensive evaluation 1)The establishment of evaluation set. Set evaluation V={v1,v2,v3,v4,v5}={good, better, generally, worse, bad}={5,4,3,2,1} 2)Statistics, determine the single factor evaluation membership vector[14-15], and formed the membership degree matrix R Membership degree in fuzzy comprehensive evaluation is the most important and basic concept. The so-called membership rij , refers to a plurality of evaluation main body to a certain evaluation object in the factor set to V assessment probability. Membership 7 vector R r , r ,, r , i 1,2,, n, r 1 ,membership ij i i1 i 2 im j 1 matrix R=(R1,R2,…,Rn)T=(rij) 3)One stage fuzzy comprehensive evaluation According to the original data to determine the solution of one stage membership fuzzy matrix RBij of each scheme, get : Bi WBi Rij 4)Two stage fuzzy comprehensive evaluation Let one stage results constitute two stage single factor judgment matrix RA [B1 B2 B3 B4 B5], A WA RA 5)Calculation of comprehensive score E AV T According to the comprehensive score height can determine various schemes, so as to select an optimal supplier. VI Summary Considering the fuzzy clustering algorithm can't distinguish between the data itself attribute imbalance, based on this algorithm with weighted fuzzy clustering and comprehensive evaluation method, the method of electronic commerce environment to select suppliers, this method can various factors taken into account, the objective reaction actual situation, the classification of the real problem can be more accurate. References [1] Ma Hongyan, Zhang Guangming.Supply Chain Distributor Performance Fuzzy Clustering Analysis[J].Business Research,2009(10):91-93 [2] Gan Ming, Wang Feng.Based on the Fuzzy Clustering Method in the Performance Evaluation of Emergency Logistics[J].Technology and Method, 2010(03):75-77 [3] Wu Nan. Application of Weighted Fuzzy Clustering in the third Party Reverse Logistics Provider Selection [J].Enterprise Forum,2010(09):33-35 [4] Schingnar, A.P. Measuring Productive Efficieney of Public Service Porvision. Fels Discussion Paper No.143,Univesity of Pennsylvania , School of Publication Ubran Policy 1980.9:20-29 [5] Huo Zhenjia.Enterprise Integrated Supply Chain Performance Evaluation Innovation and Its Assessment[M].Hebei people's publishing house,2001.8:10-15 [6] Azzone,G., Rangone,A. Measuring Manufacturing Competence:A Fuzzy Approach[J].International Journal of Production Research, 1996, 34(9):2517-2532. [7] Chen Ao. Research on Supplier Selection under the Environment of Electronic Commerce[D].Dalian University of Science and Technology.2005 [8] Wang Yingluo.System Engineering[M].Mechanical industry press,2003.7 [9] Zhu Yuxian,Wang Chengzhong,Zhang Kuiyuan, Yang Yinsheng.Fuzzy Mathematics Method[M].Chang Chun:Jilin university press.1994 [10] Manoj Kumar., Perm Vrat, Shankar,R.A Fuzzy Programming Approach of Vendor Selection Problem in Supply Chain[J]. Computer & Industrial Engineering, 2004(46):69-85 [11] Xie Jijian,Liu Chengping.Fuzzy Mathematics and Application[M].Huazhong University of Science and Technology press,2005.12 [12] Wang Guowei, Yan Li.A Weighted Spatially Fuzzy Dynamic Clustering Algorithm[J]. Computer engineering and Applications,2010(17):146-149 [13]Wu Yingxue, Long Aixiang. Application of Multistage Fuzzy Comprehensive Evaluation in the Site Selection of Logistics Center[J].Forest Engineering,2004;(05):16-19 [14]Zhang Jingzhe, Zheng Wenrui.Application of Fuzzy Dynamic Classification Method to Division of Agricultural Economic.China Science and Technology Paper Online,2002 [15]Joe Zhu. A Buyer-Seller Game Model for Selection and Negotiation of Purchasing Bids: Extension and New Models [J].European Journal of Operational Research, 2004(154):150-156